Piecewise Self-Adaption Weighted attention for the detection of concentrated distributions of ships in SAR images
Pengfei Guo, Turgay Çelik, Nanqing Liu, Heng-Chao Li · Remote Sensing Letters · 2024
Developing a robust algorithm for the detection of concentrated distributions of ships in Synthetic Aperture Radar (SAR) images presents significant challenges. These challenges stem from the intricate backscattering properties of ships with various sizes and the ambiguous and fuzzy boundaries within feature maps, especially when ships are in close proximity. To address these issues, we introduce a novel parameter-free Piecewise Self-Adaptation Weighted (PSAW) attention module, which is designed to mitigate semantic ambiguities and enhance the clarity of feature boundaries. The PSAW module comprises two distinct components: Frontal Channel Weighted Module (FCWM) and Tube Channel Weighted Module (TCWM). The FCWM is tailored to improve the discernment of salient features, while the TCWM focuses on reducing feature misalignment and incorrect positioning. By employing a self-adaptation weighting approach, the PSAW module effectively categorizes and weights the features in each channel according to different hierarchies. The effectiveness of our approach is demonstrated through extensive experiments on SSDD and HRSID datasets.